July 20, 2018

Motivating Problem

An "under-recognized" disease

Pulmonary Sarcoidosis

  • A chronic, interstitial lung disease, characterized by the presence of granulomas in the lung
  • Unknown etiology and prognosis
  • Limited treatment options

Past work

  • The Scadding staging system for sarcoidosis, based on the visual assessment of chest radiography, poorly predicts disease progression

  • An Exploration of Spatial Radiomic Features in Pulmonary Sarcoidosis (Ryan et al. 2018):
    • Disease is most apparent in the superior and lateral slices of the lungs
    • Radiomic measurements explain more information about lung function than visual assessment

Current work

  • Using radiomics, can we create disease subtypes that better predict disease progression for sarcoidosis?

  • Can we identify 3D regions of the lung where disease exists?

  • Can we compare disease across lungs between different subtypes?

Current work

  • Using radiomics, can we create disease subtypes that better predict disease progression for sarcoidosis?

  • Can we identify 3D regions of the lung where disease exists?

  • Can we compare disease across lungs between different subtypes?

Image Processing Pipeline for Lung CT

Image Processing Pipeline for Lung CT

  • Challenges:
    • EXACT09 (airways)
    • VESSEL12 (blood vessels)
    • LOLA11 (left/right lungs)
    • AAPM Thoracic (multiple organs)
  • Methods/Results (Mansoor et al. 2015):
    • Thresholding-based: Fast, easy, accurate for healthy lung
    • Region-based: Better segmentation for airways/vessels
    • Shape-based: Good for abnormal lung
    • Machine learning-based: Accurate, but computationally intensive

  • Software (Alnaser et al. 2016):
    • Pulmonary Toolkit (PTK)
    • TurtleSeg
    • Medical Imaging Interaction Toolkit (MITK)
    • ITK Snap
    • Chest Imaging Platform
    • 3D Slicer
    • MATLAB
    • R (?)

  • Challenges: EMPIRE10 (Murphy et al. 2011)

  • Registration Method: Symmetric Normalization (SyN)

  • Registration object:

  • Software: ANTs (& ANTsR), elastix (& simpleElastix)

  • Methods for Template Creation
    • Average Warped Images
    • Apply Average Transformation to Fixed Image (Li et al. 2012)
    • Apply Average Transformation to Average Warped Image (Avants et al. 2010)
  • Software:
    • ANTs buildTemplate
    • R (?)

Result: Population-level mask in initial template space

Result: Initial template mask in population-level space

Result: Population-level mask in population-level space

Developing an
Image Processing Pipeline
for Lung CT
in R

Segmentation in lungct R package

Method: Thresholding based (Mostly)

  1. Identification of lung and airways by thresholding and connected components.

  2. Elimination of airways from part (1) by thresholding, connected components, and dilation.

  3. Identification of left/right lungs by centroid of connected components, erosion and dilation.

Segmentation in lungct R package

library(ANTsRCore)
img = antsImageRead("img.nii.gz")

library(lungct)
mask = segment_lung_lr(img, lthresh = -300)

Comparison of lungct and 3D Slicer

Data: Non-smoking healthy controls from COPDGene (n=98)

Results: Dice Similarity Coefficient across all masks

  • Mean: 0.987 (Min: 0.981 - Max: 0.991)

Lung Registration in ANTsR

library(ANTsRCore)

# Read in registration objects
fixed_mask = antsImageRead("fixed_mask.nii.gz")
moving_mask = antsImageRead("moving_mask.nii.gz")
moving_img = antsImageRead("moving_img.nii.gz")

# Register objects
reg = antsRegistration(fixed = fixed_mask,
                       moving = moving_mask, 
                       typeofTransform = "SyN",
                       outprefix = "transformations")

# Apply transformation to image
warped_img = antsApplyTransforms(fixed = fixed_mask,
                                  moving = moving_img,
                                  transformlist = reg$fwdtransforms,
                                  interpolator = "linear")

Template Creation in R function: get_template, available in lungct

Based on ANTs buildTemplate

Results: Standard Template

Standard Template

Data: Right-lung, masked CT scans from non-smoking, healthy COPDGene patients (n=10)

Method: Ten iterations of get_template

Results:

  • Standard template mask
  • Mean HU of transformed scans
  • SD HU of transformed scans

Convergence to a standard shape

Sensitivity Analysis: Selection of intial template

Caution: Non-centered origin

library(ANTsRCore)
img = antsImageRead("img_noncentered.nii.gz")
antsGetOrigin(img)
## [1] -158.68555  -38.81445 -295.00000

Conclusion

Conclusion

  • Developing a standard image processing pipeline for lung CT is difficult due to:
    • High deformability of the lungs
    • Lack of unified software platforms
  • In its current stage, lungct provides a pipeline in R with basic tools for image processing
    • Segmentation of left/right lungs
    • Template creation (based on ANTs buildTemplate)
    • Registration (based on ANTsR antsRegistration)

Work in Progress

  • Creation of standard lung template (& its public release), using the entire cohort of COPDGene healthy controls

  • Motivating Problem: Perform 3D, whole lung analysis on CT scans of patients with sarcoidosis

    • Find location and severity of disease
    • Compare differences between subgroups
    • Better understand disease manifestations and prognosis

Acknowledgements

Department of Biostatistics, Johns Hopkins University

  • John Muschelli, PhD
  • Ciprian Crainiceanu, PhD

Department of Biostatistics, Colorado School of Public Health

  • Nichole Carlson, PhD
  • Debashis Ghosh, PhD

References

Alnaser, A., Gong, B., & Moeller, K. (2016). Evaluation of open-source software for the lung segmentation. Current Directions in Biomedical Engineering, 2(1), 515-518.

Avants, B. B., Yushkevich, P., Pluta, J., Minkoff, D., Korczykowski, M., Detre, J., & Gee, J. C. (2010). The optimal template effect in hippocampus studies of diseased populations. Neuroimage, 49(3), 2457-2466.

Li, B., Christensen, G. E., Hoffman, E. A., McLennan, G., & Reinhardt, J. M. (2003). Establishing a normative atlas of the human lung: intersubject warping and registration of volumetric CT images. Academic radiology, 10(3), 255-265.

Mansoor, A., Bagci, U., Foster, B., Xu, Z., Papadakis, G. Z., Folio, L. R., … & Mollura, D. J. (2015). Segmentation and image analysis of abnormal lungs at CT: current approaches, challenges, and future trends. RadioGraphics, 35(4), 1056-1076.

Murphy, K., Van Ginneken, B., Reinhardt, J. M., Kabus, S., Ding, K., Deng, X., … & Vercauteren, T. (2011). Evaluation of registration methods on thoracic CT: the EMPIRE10 challenge. IEEE transactions on medical imaging, 30(11), 1901-1920.

Ryan, S. M., Fingerlin, T., Hamzeh, N., Maier, L., & Carlson, N. (2018). An Exploration of Spatial Radiomic Features in Pulmonary Sarcoidosis. arXiv preprint arXiv:1806.10281.

Contact